- Analysis of multiple groups of images shows these thresholds to be quite robust for our system. They may not hold for all crop image analysis problems.
Typical Threshold Values
| Class | Color Threshold | Bright |
|---|---|---|
| Green plants | Low red:green ratio (<~1.0) |
Intermediate-High (~ 200 - ~ 600) |
| Purple Leaves and Flowers | Low green:blue ratio (< ~ 1.0) |
Intermediate-High (~ 200 - ~ 600) |
| Plant Residue | Intermediate-High red (> ~ 180) |
Higher (> ~ 430) |
| Bare Soil | Lower red (< ~ 160) | Lower (< ~ 400) |
- We can use these typical threshold values to develop a classification scheme to classify all pixels in an image (Figure 3).
Figure 3. Pixel Classification Scheme
- We wrote a program using CIMG (a free C++ library) that opens an image as a matrix of red, green, and blue color values, and classifies pixels based on threshold values provided. The program returns pixel counts for the leaves (green and purple), residue, and soil as well as ambiguous (unclassified) pixels and pixels within 2% of a threshold (borderline). The program also generates a false-color image to allow evaluation of accuracy of classification (Figure 2F, 3B).
